Heritability of Social Status

Clark says that social scientists frequently underestimate the heritability of social status. His argument is based on a well-known bias in the statistical measure known as correlation. For any quantitative characteristic, such as height, income, or IQ, we assess heritability by measuring the correlation between the observed values of parents and the observed values of their children. The correlation can vary between negative one and one. Negative one would mean perfect inverse correlation—a characteristic that is high for parents is low for children, and vice-versa. Positive one would mean perfect (positive) correlation, so that children always inherit all of this characteristic from their parents. A correlation of zero would mean that the children inherit none of the characteristic from their parents, so that its values are determined entirely by environmental and random factors.

The inherent bias in using correlation is that whenever measurement error is present, correlation is biased toward zero. To the extent that measurement error matters, investigators who measure heritability by looking at the correlation between parents and children will underestimate heritability.

For example, suppose we were trying to measure the heritability of skill at playing poker. We could measure the correlation between average poker winnings of parents and average poker winnings of children. However, average winnings are not a precise measure of poker skill. Other factors besides skill may affect average winnings. These include luck in getting good cards as well as the skills of the opponents one happens to be playing against. These are sources of measurement error when using average winnings as a measure of poker skill. The greater this measurement error, the more strongly the correlation will be biased toward zero, and the more serious will be the underestimate of the heritability of skill at playing poker.

Clark’s insight is that one way to correct for measurement error is to look at multi-generational regression toward the mean. A true measure of heritability should decay with the square of that measure. For example, if poker skill is 50 percent heritable in children, then it should be 25 percent heritable in grandchildren, 6.25 percent heritable in great-grandchildren, and so on. However, measurement error should be the same between any two generations, regardless of how close they are.

Now suppose that, based on the correlation between winnings of parents and children, you estimate the heritability of poker skill at 50 percent. But then you measure the correlation between parents and great-grandchildren and find that it is 20 percent, rather than the 6.25 percent that would be consistent with your estimate of heritability. This would suggest that there is a significant problem of measurement error, and the true heritability of poker skill is much higher than 50 percent.

Clark and his researchers looked at multi-generational outcomes on a variety of measures in several countries. They concluded that under many different institutional arrangements and across many time periods, the true correlation across generations in social status is somewhere between .7 and.8, which is much higher than most conventional estimates. In short, persistence of social class is much higher than most researchers believe it to be, based on single-generation correlations that are biased downward by measurement error.

Note that this bias could affect any quantitative indicator that is subject to measurement error. For example, suppose that general intelligence (IQ) is measured with error. This error could be because a person could happen to make good or bad guesses on the day that he or she takes a particular IQ test. Or the test itself could be an imperfect measurement instrument. To the extent that these sources of measurement error matter, IQ might be more heritable than thought. This would be revealed if we were to observe a higher-than expected correlation between the IQ of parents and grandchildren given the correlation of IQ between parents and children.

Measurement Error in Social Status

Social class can be measured in a variety of ways, including “income, wealth, education, occupational status, and even longevity.” (Clark, page 1) Because each of these depends in part on luck, they are all imperfect measures of the “true” skill that someone brings to the poker game of life.

For example, suppose that a man with a genetic endowment that makes him likely to live past age 90 has the misfortune of being killed in an auto accident at age 40. His measured longevity will be an erroneous measure of his biological makeup. These sorts of chance occurrences will bias downward one’s estimate of the heritability of longevity based on single-generation correlation.

Clark points out two sources of measurement error in income as an indicator of social status.

First, there is an element of luck in the status attained by individuals. People happen to choose a successful field to work in or firm to work for. They just succeed in being admitted to Harvard, as opposed to just failing. Second, people make tradeoffs between income and other aspects of social status. They may choose to be philosophy professors instead of finance executives. (page 11)

Because of these factors, single-generation correlations tend to under-estimate the heritability of social class. To put it another way, such correlations over-estimate social mobility.

Deterministic Findings

Clark writes:

This book estimates social mobility rates by measuring the rate at which surnames that originally had high or low social status lose that status connotation….

Surname status shows regression to the mean in all cases, but the process is slow. Elite surnames can take ten or fifteen generations (300-450 years) to become average in status…

[S]ocial mobility seems to occur at a similar rate for different measures of status: wealth, education, occupational status, and membership in political elites…

… the rate of persistence is close to constant across wildly different social systems. It is little higher for the feudal England of the Middle Ages than for the progressive, equality-loving, social-democratic Sweden of today. (pages 105-106)

As Clark points out, each of these findings reinforces the hypothesis that social status is highly heritable. Some people are skilled at the poker game of life, and these skills tend to be passed on to their children. Even seemingly major differences across societies in the rules of the game do not seem to affect the outcome.

Critics will argue that the persistence of social status might not be due primarily to genetic factors. Clark would reply that at a global level, the evidence that different institutional structures yield similar outcomes in terms of mobility argues in favor of genetic determinism. In addition, at a family level, he writes:

The research on adoption outcomes, however, implies strongly that most of the variation in outcomes for adopted children stems from their biological parents or from chance, not from their adoptive parents. Biology may not be everything, but it is the substantial majority of everything. However, the adoption studies do leave open the possibility that social interventions could change outcomes for children from the most disadvantaged backgrounds. (page 264)

Ambiguous Implications

For libertarians, the implications of Clark’s finding of strong heritability of social status are ambiguous. On the one hand, his findings argue against extensive efforts at social engineering that try to achieve parity across groups. Clark writes,

… the simple model of social mobility outlined here… explains the often-observed slow rates of social mobility for specific social, ethnic, racial, and religious groups without having to posit discrimination, ethnic capital, or ethnic social connections as contributing factors…

A recent study by the Pew Charitable Trust found that among families with middle-class incomes, black and Latino children were more likely to fall into the bottom third of the income distribution than white children. For whites, the chance was 25 percent, for blacks 37 percent, and for Hispanics 29 percent. (pages 123-124)

Clark suggests that this may reflect that the underlying mean for these ethnic groups may differ, and the higher propensity of middle-income blacks and Hispanics to have their children’s income fall to the bottom third might be due to regression toward a lower mean.

Consider, for example, two families with annual incomes of $90,000, one Jewish and another black. Since median household income in the United States is $52,000, we expect that the children of such families will have incomes lower than their parents’ and closer to the median. However, this income level is close to the median for the Jewish community. Thus the random component to this family’s income, the part that deviates from what we would expect from their underlying social capabilities and is not heritable, will be on average zero. So their children will regress only modestly toward the social mean. In contrast, a black family with a median income of $90,000 has nearly three times the median income of the black community, which is only $35,000. Families who have incomes well above the average of their community typically have benefited from a positive random shock to their income and that has placed them above their underlying social status… So the black children will show, on average, a greater drop in income relative to the parents than the Jewish children. (page 125)

Attempts to engineer different outcomes tend to produce perverse results. For example, Clark writes,

One factor that might be predicted to increase social mobility in India is a form of affirmative action known as the reservation system, whereby up to half of public-sector jobs and places in educational institutions are reserved for disadvantaged social groups… its main effect has been to create a new elite composed of groups that were never significantly disadvantaged. In practice, the system has largely hurt the prospects of the truly disadvantaged. (page 143)

On the other hand, his findings argue against the need to create strong incentives to succeed. If some people are genetically oriented toward success, then they do not need lower tax rates to spur them on. Such people would be expected to succeed regardless. The ideal society implicit in Clark’s view is one in which the role of government is to ameliorate, rather than attempt to fix, the unequal distribution of incomes. As Clark puts it,

If social position is largely a product of the blind inheritance of talent, combined with a dose of pure chance, why would we want to multiply the rewards to the lottery winners? Nordic societies seem to offer a good model of how to minimize the disparities in life outcomes stemming from inherited social position without major economic costs. (page 15)

I expect that Clark’s genetic determinism will trouble many readers of his book. It is important to remember that he does not believe that any individual’s fate is genetically determined. Rather, he is arguing that social outcomes in the aggregate tend to average out in a way that suggests a very strong genetic component.